- Updating......
The code is based on FCN_8s
- ISPRS Vaihingen dataset
- ISPRS Potsdam dataset
- The original dataset can be requested for the download from here.
- You should cut the training images as well as corresponding labels into patches with an overlap of 171 pixels.
- Python == 3.7.10
- PyTorch == 1.8.1
- CUDA ==10.1
For Vaihingen, run:
sh train_vaihingen_fcn.sh
For Potsdam, run:
sh train_potsdam_fcn.sh
The results will be saved in the ./snapshots/
folder.
For test, run:
sh Eval.sh
For different datasets, please manully change dataset
in Eval.sh
.